GoogleEmbeddingGemma
EmbeddingGemma 2
EmbeddingGemma 2 embeds text, code, images, video and audio in a shared vector space.
What the release provides
The 740M-parameter model combines a 270M text backbone with optional 170M vision and 300M audio encoders. It accepts 8,192 tokens and produces 768-dimensional vectors, with supported truncation to 512, 256 or 128 dimensions.Google EmbeddingGemma 2 model card
How to use it well
Select only the input modalities the application needs. Keep query and document dimensions aligned, and normalize truncated vectors before similarity scoring. Evaluate the smaller dimensions on each required modality; a text-only retrieval result does not establish image or audio search quality.Google EmbeddingGemma 2 model card
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | google/embeddinggemma-2 |